IndicRecipeNutri

A corpus of 220,187 Indian recipes with multilingual provenance and estimated dish-level micronutrition; a typed knowledge graph of 223,406 nodes and 6,270,560 edges across 17 node types and 21 relation types; a 68-query retrieval benchmark with knowledge-graph-derived gold sets; and a synthetic collaborative interaction log. The food-composition layer uses the IFCT 42-nutrient schema but contains no IFCT and no Indian composition data; its values are USDA SR Legacy and UK CoFID. Two-tier release. Structured, factual and derived fields are published under CC BY-NC-SA 4.0 with per-recipe source attribution. Raw recipe prose is not redistributed; a rehydration index of source URLs and SHA-256 text digests is published instead, together with a re-fetch script. Known defects are published, not hidden. An independent allergen and diet audit ships with the data. See the datasheet before use, and do not use the allergen flags for end-user safety decisions.